shwetanlp / chq-summ Goto Github PK

PyTorch code for ACL 2021 paper Reinforcement Learning for Abstractive Question Summarization with Question-aware Semantic Rewards

Python 100.00%

chq-summ's Introduction

Reinforcement Learning for Abstractive Question Summarization with Question-aware Semantic Rewards

The code requires Python 3 and please install the Python dependencies with the command:

pip install -r requirements.txt

The original MeQSum dataset is available here.

Running the code

Please make sure to download the pre-trained question-type identification and question-focus recognition models from here and place it in the current directory.
Fine tune ProphetNet model on MeQSum dataset

Follow the instrcution from transformers repo. https://github.com/huggingface/transformers/tree/v4.1.1/examples/seq2seq

Train MLE + RL Model

python main.py --train_mode rl --trained_model_path /path/to/the/fine-tuned/prophetnet/model

Test Model

python main.py --model test --trained_model_path /path/to/the/saved/model

Reference

If you are using this code for your reseach work then please cite our paper:

@inproceedings{yadav-etal-2021-reinforcement,
    title = "Reinforcement Learning for Abstractive Question Summarization with Question-aware Semantic Rewards",
    author = "Yadav, Shweta  and
      Gupta, Deepak  and
      Ben Abacha, Asma  and
      Demner-Fushman, Dina",
    booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers)",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.acl-short.33",
    doi = "10.18653/v1/2021.acl-short.33",
    pages = "249--255"
    }

chq-summ's People

Contributors

Stargazers

Watchers

Forkers

dongguanting wzy232303 iman998 banhkun zurichrain jatins-dev

chq-summ's Issues

Memory grows with iterative training

Hi, when i train the model, I found that with iterative training, the memory of GPU keeps increasing. And at last, it returns an error: Out of Memory.
I use the V100, 32G memory, it is weird.

About the sample sequence length

Can you send me the Fine tune ProphetNet model or tell me the specific steps??

Hi, I just want to run the whole code to see the specific process, but I have a problem in the second step, can you send me the fine-tuned ProphetNet model to see or tell me the specific steps? Thans very much

Issue with pre-trained question-type identification and question-focus recognition models

@shwetanlp

Steps to reproduce the issue

1.Complete the step 1 & 2 given in the readme.md of the repository
2. Download the pre-trained model in the current directory
3. Run the command python main.py --train_mode rl --trained_model_path /path/to/the/fine-tuned/prophetnet/model

Expected result

It should train the model

Actual result

python main.py --train_mode rl --trained_model_path /home/sosukeaizen/Desktop/saved_model/prophetnet-meqsum-model

Some weights of the model checkpoint at bert-base-uncased were not used when initializing BertForSequenceClassification: ['cls.predictions.transform.dense.weight', 'cls.seq_relationship.weight', 'cls.predictions.transform.dense.bias', 'cls.seq_relationship.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.decoder.weight', 'cls.predictions.bias', 'cls.predictions.transform.LayerNorm.bias']
- This IS expected if you are initializing BertForSequenceClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing BertForSequenceClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of BertForSequenceClassification were not initialized from the model checkpoint at bert-base-uncased and are newly initialized: ['classifier.bias', 'classifier.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
Traceback (most recent call last):
  File "main.py", line 17, in <module>
    from reward.compute_question_type_reward import get_question_type_reward
  File "/home/sosukeaizen/Desktop/task/CHQ-Summ/reward/compute_question_type_reward.py", line 67, in <module>
    type_model.load_state_dict(torch.load('type_model/epochs-7'))
  File "/home/sosukeaizen/.local/lib/python3.8/site-packages/torch/serialization.py", line 579, in load
    with _open_file_like(f, 'rb') as opened_file:
  File "/home/sosukeaizen/.local/lib/python3.8/site-packages/torch/serialization.py", line 230, in _open_file_like
    return _open_file(name_or_buffer, mode)
  File "/home/sosukeaizen/.local/lib/python3.8/site-packages/torch/serialization.py", line 211, in __init__
    super(_open_file, self).__init__(open(name, mode))
IsADirectoryError: [Errno 21] Is a directory: 'type_model/epochs-7'

System information (as much as possible)

Ubuntu 20.04.5 LTS

Additional comments

It is not able to identify type_model/epochs-7'

About MATINF

I want to reproduce and compare your results on MATINF, can you provide your processed(translated) version of MATINF?

About the dataset ?

In the main.py. Line 288-289 and 306-308。
is train dataset equals to the val dataset ? Is that right?